A random arrival is more likely to land in a long gap than a short one.
Operations & Systems.
Queues, bottlenecks and feedback in the systems we build.
31 live experiment · 0 plannedAn extra shortcut can make selfish routing slower for everyone.
A part that cannot be sped up limits the gain from adding more workers.
More items in a stable process usually means a longer time spent inside it.
Small customer changes can become larger swings in upstream orders.
A project can take longer than the inside-view estimate suggests.
A nearby, larger target is generally easier to reach.
Too many separate permissions can leave a useful resource unused.
A smaller inflow can still leave a stock growing.
Adding a seat can make one group lose its allocation.
Watch a concentration spread while the total stays constant.
Yesterday’s price can make tomorrow’s market oscillate.
Compare a chain of dependencies with independent backups.
The same input can leave a switch on or off depending on its history.
Route traffic and compare individual incentives with total delay.
Commit an investment before bargaining and inspect whose incentives cover its cost.
Steer toward a target using delayed measurements and discover overcorrection.
Spend three acceleration tokens on a six-task project and try to launch sooner.
Report a benefit, fund a project and inspect the pivot payments.
Compare unit costs when wages rise faster than one sector's productivity.
Spend six resource tokens to lift a fictional plant’s limiting resource.
Trace tentative distances through a weighted network until the cheapest route is certified.
Draw walls and compare grid searches using different heuristic weights.
Replay the same page requests through three-slot and four-slot caches under FIFO and LRU.
Add a server to a hash ring and inspect exactly which keys move.
Reveal independent arrivals and compare random gaps with the expected event rate.
Order a perishable stock quantity before revealing tomorrow’s uncertain demand.
Balance annual ordering cost with the cost of holding a larger average inventory.
Hold utilisation fixed while changing arrival and service variability.
Double cumulative production and inspect an assumed unit-cost progress ratio.
Queue five known-duration jobs and compare complete schedules’ average waiting.